AI Visibility to Local Leads

AI-powered search is beginning to influence how customers discover local businesses, but measuring exactly how many leads come from these platforms remains difficult.

CallRail says AI citation clicks now account for around 1% to 2% of calls generated for its customers. Sean McCrohan, the company’s VP of Technology, said during SEJ Live on 26 August that this figure has roughly doubled since January.

McCrohan and local search specialist Steve Wiideman discussed how AI assistants are increasingly involved in the customer journey, while existing reporting tools still struggle to capture the full picture.

AI Referrals Are Growing

Customers do not necessarily move directly from an AI recommendation to a phone call. Instead, they may use an AI assistant for initial research before carrying out a branded Google search or visiting a business website.

This makes it difficult to identify exactly where the original lead came from.

At present, two forms of AI attribution can be tracked relatively easily.

The first is a customer clicking a citation included in an AI-generated response and subsequently visiting the business website. The second is when a customer directly tells the business that an AI platform, such as ChatGPT, recommended it.

Call recordings can also reveal cases where customers mention an AI recommendation despite there being no measurable referral click.

Both measurable methods have grown by more than 100% since January, although they started from a relatively small base.

For the multi-location and franchise businesses Wiideman works with, filtering Google Analytics 4 traffic from platforms including ChatGPT, Gemini, Claude, Perplexity and Grok produces a figure close to 1%.

Rather than viewing this as a direct sales channel, Wiideman considers AI primarily a discovery mechanism that can influence customers before they contact a business.

AI-Driven Leads Can Arrive Around the Clock

The timing of AI-influenced traffic is another consideration for local businesses.

McCrohan said CallRail has historically found that roughly half of website traffic takes place outside normal business hours. For traffic associated with AI search, that figure rises to almost two-thirds.

Customers may begin researching a service during the afternoon and continue looking for information late into the evening.

This creates an opportunity, but also a challenge. An AI assistant may recommend several businesses rather than a single provider, meaning a potential customer could easily move on to another company if the first one does not answer.

For businesses, having a way to handle enquiries outside traditional opening hours could therefore become increasingly important.

Tracking AI Phone Numbers Requires Server-Side Changes

Businesses looking to attribute calls from AI crawlers also need to understand how these systems access websites.

McCrohan said CallRail has found that server-side number replacement can work when a separate telephone number is displayed to AI crawlers.

However, browser-based number swapping is less effective because many AI crawlers do not execute the JavaScript used to make those changes.

This means businesses relying on client-side scripts to replace phone numbers may not see the same results when their websites are accessed by AI agents.

Wiideman also warned businesses to be careful when using user-agent detection. Showing substantially different information to different crawlers can create potential cloaking concerns.

For local businesses, keeping the same canonical name, address and telephone information across important platforms remains essential. These details need to be consistent with information used by Google and other local search services.

AI Prompt Tracking Should Focus on Trends

Wiideman recommends that businesses develop a library of prompts based on the claims and topics they want AI systems to associate with their brand.

One approach is to identify specific claims that accurately describe the business and search for them to see which other websites make similar statements.

These can then form a prompt library of roughly 100 to 125 searches that can be monitored over time.

However, businesses should avoid treating the results as fixed rankings.

AI platforms can produce different answers to the same question, sometimes even when the same person repeats the prompt. This phenomenon, sometimes referred to as prompt drift, makes individual results less reliable as a measurement.

Analysis cited during the discussion found that repeating the same local search in Gemini produced overlapping cited sources around 40% of the time. The same business appeared as the top recommendation only around 7% of the time.

That is very different from traditional Google local results, where the leading business was reported to remain consistent around 90% of the time.

The implication is that businesses should look for broader patterns rather than becoming too focused on whether a particular website is cited for an individual prompt.

Customer Conversations Can Reveal Better Keywords

Businesses may already have a valuable source of information about how customers search for their services.

Call recordings and transcripts contain the actual language people use when describing their problems. Those phrases may differ considerably from the terminology used on the company’s website.

McCrohan argues that identifying these differences can help businesses understand what customers are really looking for and where their content may not match customer needs.

Website chat conversations can provide similar insights. Customers often begin by describing their problem in their own words through a chat tool, yet businesses do not always analyse these conversations.

Reviewing calls and chat logs could therefore help create more relevant content and expand the prompt library used to monitor AI visibility.

Reviews Still Influence AI Recommendations

Traditional reputation signals have not disappeared simply because AI search is becoming more important.

Wiideman recommends that businesses with multiple locations encourage customers to leave reviews across the platforms they already use, rather than concentrating entirely on Google Business Profile.

Platforms such as Reddit, TripAdvisor and Yelp can also influence how businesses are represented online.

For local businesses, maintaining a strong and consistent reputation across several relevant platforms can therefore support visibility beyond traditional search.

Centralised Data Becomes More Important

Managing accurate information becomes increasingly difficult as the number of business locations grows.

Wiideman says businesses with dozens or even thousands of locations need centralised control over key information, including schema, data feeds and telephone numbers.

Without this oversight, individual locations can develop inconsistent listings or introduce information that conflicts with the wider brand.

For larger organisations, maintaining accurate business data may now be just as important as traditional ranking factors when it comes to appearing in AI-generated recommendations.

AI Search Could Generate Leads Without a Website Visit

The role of a website may also change as AI assistants become more capable.

In the future, an AI agent could identify a suitable business, verify its contact information and potentially offer to place a call on the customer’s behalf.

In that situation, the customer might never visit the company’s website at all.

That does not necessarily mean the website has failed. It may have provided the information that allowed the AI system to recommend the business in the first place.

For local businesses, the focus is therefore beginning to shift from simply tracking website traffic to understanding whether accurate online information is helping generate real-world enquiries.

AI search is still developing, and the available attribution data remains limited. However, the early figures suggest that AI referrals are growing, making it increasingly important for businesses to monitor how they are being discovered and recommended.